{"paper":{"title":"Constrained Policy Optimization with Cantelli-Bounded Value-at-Risk","license":"http://creativecommons.org/licenses/by/4.0/","headline":"VaR-CPO approximates Value-at-Risk constraints via Cantelli's inequality to guarantee zero violations during training in feasible reinforcement learning environments.","cross_cats":["stat.ML"],"primary_cat":"cs.LG","authors_text":"Jan-Peter Calliess, Rohan Tangri","submitted_at":"2026-01-30T13:57:47Z","abstract_excerpt":"We introduce Canary, a risk-averse method designed to optimize Value-at-Risk (VaR) constrained reinforcement learning (RL) problems. We employ Cantelli's inequality to obtain a tractable, conservative and smooth bound on the VaR constraint based on the first two moments of the cost return. This yields a constraint estimator that remains stable with tight violation thresholds in dense cost regimes. Extending the trust-region framework of the Constrained Policy Optimization (CPO) method, we further provide worst-case bounds for both policy improvement and constraint violation during the training"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"VaR-CPO achieves zero constraint violations during training in feasible environments, a property baseline methods fail to uphold, while providing worst-case bounds for policy improvement and constraint violation.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The Cantelli inequality provides a sufficiently tight and conservative approximation to the true VaR constraint so that enforcing the bound still guarantees the original probabilistic constraint in practice.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"VaR-CPO approximates non-differentiable VaR constraints via Cantelli's inequality to enable safe, sample-efficient policy optimization with zero training violations in feasible environments.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"VaR-CPO approximates Value-at-Risk constraints via Cantelli's inequality to guarantee zero violations during training in feasible reinforcement learning environments.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"5833deae7b8bf68db5033edbd4f3443af94e8f792a74de4f7a049cff80bd6916"},"source":{"id":"2601.22993","kind":"arxiv","version":4},"verdict":{"id":"7b84fecb-eea6-421d-90b3-a8385299ea5b","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-16T09:35:17.857407Z","strongest_claim":"VaR-CPO achieves zero constraint violations during training in feasible environments, a property baseline methods fail to uphold, while providing worst-case bounds for policy improvement and constraint violation.","one_line_summary":"VaR-CPO approximates non-differentiable VaR constraints via Cantelli's inequality to enable safe, sample-efficient policy optimization with zero training violations in feasible environments.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The Cantelli inequality provides a sufficiently tight and conservative approximation to the true VaR constraint so that enforcing the bound still guarantees the original probabilistic constraint in practice.","pith_extraction_headline":"VaR-CPO approximates Value-at-Risk constraints via Cantelli's inequality to guarantee zero violations during training in feasible reinforcement learning environments."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2601.22993/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"9040767e148b8c72e9fc099919b372291776c02fefffa41b70a4c425745a302e"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}